I’m now a full-time software developer in Mozilla Labs working primarily on Weave [labs.mozilla.com]. I suppose that means I’ll have to pay income taxes again.. 😉 but that’s insignificant compared to helping Weave redefine web clients where users can interact with websites and friends in whole new ways. I’ll have less time to write about the EdBrowser, but being in Labs, I might get to build some of this as a bunch of ideas for that web client was built around Weave. 🙂

Most of what I worked on during the week was getting familiar with the Weave codebase and some initial debugging/hacking. I also did a little bit on optimizing the Location Bar for Firefox 3.1+.

Is 4 > 20? Yes! ?

My first task was figuring out why 4 is larger than 20 [bugzilla.mozilla.org]. This was causing problems with Weave history sync on Fennec when we set the Places sorting mode to get a number of recent pages. So to try figuring out what went wrong, I started with my Unibody Macbook and installed VMWare Fusion [vmware.com] to run Ubuntu 8.10 [ubuntu.com] for a cross-compiling scratchbox [wiki.mozilla.org] to build Fennec [wiki.mozilla.org] for the Nokia N810. (Side note: don’t run hg from the scratchbox. It’s old. Like 0.9.1 old.)

That’s just to build Fennec. Getting it to run on the Nokia N810 is just as fun. The default package is bzipped, but that doesn’t work on the device by default, so I have to repackage it as gzip. Even before then, you’ll need to jump through hoops to get Fennec onto the device, but the easiest way seems to be putting stuff on a web server and downloading it wirelessly to the device.

Long story short, I used my x86 asm experience from doing computer architecture/compiler research at UIUC, and noticed the compiled code for SetSortingMode [hg.mozilla.org] was doing a plain comparison of the input argument to the value 20.

cmp r1, #20

While the value 4 was getting passed into the unsigned short argument, the compiler was expecting a 4-byte sign extended value. Vlad gave me a quick patch that made the XPTCall magic handle smaller-than-4-byte types correctly, and all was good. 4 is now less than 20.

Initial Weave Hacking

I got a Weave account and used it for the first time. (And so can you!) [services.mozilla.org] I kept track of various small UI issues as I used the Firefox client and pushed fixes for them — small localized changes are a great way to get into the bigger project. But I eventually messed up my account by typing different passphrases from different computers.

Lucky for me, I’m working on a way for users to reset the data on their client/server [bugzilla.mozilla.org]. The general idea here is to allow the user to “start over” incase something went wrong. 1) Clean sync with the server, 2) Restart the client from server state, 3) Make other clients like the current one, 4) Have clients repopulate the server.

Before I dove too deep into coding, I got a whole look at the Weave architecture to better understand what pieces od what: A bird’s-eye view [toolness.com] and Client APIs [wiki.mozilla.org]. Also, Weave is written with a pattern called Trampolining Generators [neilmix.com]. This basically lets you use asynchronous calls synchronously and get something more like traditional threading.

Color-coded log files for easier tracking of Weave behavior

I also figured I would be looking at debug logs a bunch, so I decided to spice up the plain black/white log file [bugzilla.mozilla.org] and color code the text by the type of message (warning, info, debug, etc.) and the background by the component generating the message. Hopefully now I’ll be able to get deep into Weave hacking.

AwesomeBar Speedup

I posted about speeding up the location bar, and it’s made it into Firefox 3.2 already. On the way there, I used Andrew Sutherland’s SQLite explain parser to better understand why things were slower and faster. Basically, a query becomes a set of opcodes like how javascript code becomes an intermediate bytecode, and the engine executes this stream of instructions.

Andrew first made this tool to analyze his own query optimizations [visophyte.org], and now he’s made the code available for everyone to use and contribute [hg.mozilla.org]. Thanks!

It seems that Mercurial and Mozilla have quite different patch/review process, but then again they’re two very different projects. I just found out today that Bug 394650 – Make line numbers linkable when viewing files/annotate for hg.mozilla.org [bugzilla.mozilla.org] was “works for me.” I was confused for a little bit; but yes indeed, annotate/changeset views on hg.mozilla.org have linkable line numbers for each file… just like how I would have coded it. 😉

Turns out that the patch that I sent in to the dev list ended up making it in to Mercurial 1.0 [hg.intevation.org]. No official reviews, no comments. They just checked in the patch to their “crew” (trunk) repository and let it bake for several months. Seems like nobody complained enough to have it backed out and here it is now in Mercurial 1.0 as we can see on hg.mozilla.org.

But this post isn’t about the differences between Mercurial and Mozilla. It’s about the current Mozilla CVS development process vs Mozilla with Mercurial.

One main difference is changesets. Instead of having a separate version number for each file as in CVS, all related changes are grouped together for a changeset. These are the steps one needs to do right now to see all related changes of a patch: 1) use mxr to find the file you’re editing 2) switch to blame to find the checkin 3) click the bug number if provided 4) hope there’s a clearly marked patch as the one checked in to see what else was changed. 🙁

With Mercurial, you can just look at the changeset which contains everything; e.g., patch to not need to click the page to use gmail keyboard commands [hg.mozilla.org]. Additionally, now with my patch for showing line numbers in changesets, you can link people to a specific line in a patch [hg.mozilla.org]. This could be useful if you need to point to a particular line in a bugzilla comment about why it broke something. Or especially useful with a stack of patches on mq and you’re collecting feedback on it from other people.

Another aspect of changesets is that there’s a “global version” for the whole repository. You can jump back in time to an earlier revision and view the whole repository as it was before a particular patch was checked in. We currently have something like this on a per-file basis with bonsai, but you need to do so much more work to look at “the right version” of other files. E.g., going back to when adaptive learning was checked in [hg.mozilla.org], you can see that the autocomplete was still only emphasizing the first match in the url and title [hg.mozilla.org].

Of course, you can just use the mercurial web interface to view the latest version of the file. Just take http://hg.mozilla.org/cvs-trunk-mirror/index.cgi/file/tip/ and append the file you want. (Normally “tip” would be a changeset/revision hex number, but you can put in “tip” because it’s an alias (tag) for the latest version.) And from there, you can look at the whole patch for the current version (changeset), browse the checkin history of that file (revisions), or look at the blame (annotate) [unfortunately it doesn’t have checkin comments on hover yet].

Typing out that whole url might be cumbersome, but good thing we have the AwesomeBar with multi-word search on word boundaries (with camel case) plus adaptive learning. 😉 “hg. auto” [hg.mozilla.org]

I’m not sure how distributed the development process will become for Mozilla, but these neat tools provided by default with Mercurial will make it easier to track changes and refer to them in bugzilla comments or over IRC. (Oh, and if you’re using mercurial and haven’t looked into mq yet, you should do so. It’s a great (builtin) extension to track/reorder multiple patches that you’re working on. And for those curious how I send changes for try-server builds, I apply the stack of patches and hg export firstpatch:lastpatch > combined.patch 😀 )